Faculty Research Working Papers Series
Crossing the Water’s Edge: Elite Rhetoric, Media Coverage
and the Rally-Round-the-Flag Phenomenon, 1979-2003
Matthew A. Baum
John F. Kennedy School of Government - Harvard University
Tim Groeling
University of California, Los Angeles
March 2007
RWP07-013
The views expressed in the KSG Faculty Research Working Paper Series are those of the author(s) and do not
necessarily reflect those of the John F. Kennedy School of Government or of Harvard University. Faculty Research
Working Papers have not undergone formal review and approval. Such papers are included in this series to elicit
feedback and to encourage debate on important public policy challenges. Copyright belongs to the author(s).
Papers may be downloaded for personal use only.
Crossing the Water’s Edge: Elite Rhetoric, Media Coverage and the Rally-Roundthe-Flag Phenomenon, 1979-2003.
The most widely accepted explanation for the rally-round-the-flag phenomenon is a relative
absence of elite criticism during the initial stages of foreign crises. In this study we argue that the
nature and extent of elite debate may matter less than media coverage of any such debate, and that such
coverage is heavily influenced by commonly held professional incentives and norms that lead
journalists to strongly prefer certain stories over others. We also argue that not all messages in this
debate matter equally for public opinion. Rather, the persuasiveness of elite messages depends on their
credibility, which, in turn, arises out of an interaction between the sender, receiver, and message.
Hence, only by understanding the interactions between elites, the public, and the press can we account
for variations in public responses to presidential foreign policy initiatives. We test our theory by
examining public opinion data and network news coverage of all major U.S. uses of military force from
1979 to 2003. We content analyze all congressional evaluations of the president and the executive
branch of government from the three network evening newscasts within 60-day time periods centered
on the start date of each use of force. Our results offer strong support for the theory.
Matthew A. Baum
Harvard University
John F. Kennedy School of Government
(on leave from UCLA)
[email protected]
Tim Groeling
University of California, Los Angeles
[email protected]
1
In August 2005, senators Chuck Hagel (R-NE) and George Allen (R-VA) appeared together on
ABC’s This Week to discuss the current and future status of American involvement in Iraq. The senators
were of comparable stature; both were considered credible aspirants for the 2008 Republican
presidential nomination, both were forceful and articulate for their respective positions, and both spoke
for similar lengths of time. Yet in the two weeks following the interview, journalists broadcast over 20
times as many television stories about Hagel’s criticism of the war than about Allen’s defense of it.1 In
this study, we argue that the differential coverage of these prominent Republicans was both predictable
and representative of an important limitation in our understanding of the dynamics of public support for
the president, especially in times of foreign policy crisis.
Scholars have long debated the causes and consequences of public support for the overseas
application of military force (e.g., Lippmann 1934, Almond 1950, Rosenau 1961, Baum 2003, Holsti
2004, Eichenberg 2005, Page and Bouton 2006). To explain public support, research in this area has
focused on the characteristics of the conflicts themselves (hereafter “event-based” explanations), the
internal characteristics of individual citizens (“individual-level” explanations), or on the domestic
political circumstances surrounding them (“domestic political” explanations).
Event-based explanations focus primarily on longer-term public support, arguing that a
president’s ability to sustain public support for a U.S. military engagement depends primarily on its
degree of success (Kull and Ramsay 2001, Feaver and Gelpi 2004), or the number of U.S. casualties
(Mueller 1973, Gartner and Segura 2000). Such explanations cannot, at least in many instances, account
for the presence or absence of a public opinion rally at the outset of a military conflict, before the public
observes either the ultimate costs or outcome (for a critique of these literatures, see Berinsky 2005).
Jentleson (1992), however, advances an event-based theory that can, at least potentially, account
for both initial and longer-term public support for U.S. conflicts. He argues that the American public is
more likely to support military actions perceived as defensive (aimed at imposing “foreign policy
2
restraint” on an adversary), rather than offensive (aimed at imposing “internal political change”) in
nature (see also Oneal et al. 1996, Jentleson and Britton 1998, Eichenberg 2005). Yet recent research
into both the rally-round-the-flag phenomenon (e.g., Brody 1991, Baum 2002) and, more generally, the
framing of foreign policy (e.g., Entman 2004) calls this argument into question. Such scholarship has
shown that public perceptions concerning the offensive or defensive nature of U.S. military
engagements are often endogenous to the domestic political circumstances surrounding them, including
the efforts of elites to frame events to their own advantage (Entman 2004).
Presidents routinely seek to frame their military actions as self-defense (e.g., Baum 2003, Perla
2005). At the individual level, most Americans know relatively little about foreign affairs (Holsti 2004).
Consequently, in determining whether to support or oppose a conflict, typical Americans are ill equipped
to independently assess the President’s “true” motivations, especially in the short-term. Instead they rely
on information shortcuts, or heuristic cues (Sniderman et al. 1991, Popkin 1994), most notably the
opinions of trusted political elites, primarily as reflected in the mass media. Trust, in turn, frequently
hinges on one particularly accessible heuristic: party identification (Rahn 1993, Popkin 1994, Nelson and
Garst 2005). 2 Individuals’ interpretations of heuristic cues depend in significant measure on their preexisting belief systems (Hurwitz and Peffley 1987, Herrmann et al. 1997), for which party identification is
typically an important (Rahn 1993, Popkin 1994, Lupia and McCubbins 1998, Groeling 2001, Nelson and
Garst 2005), albeit incomplete (Holsti 2004), element. The party affiliations of information sources (e.g.,
elites) and receivers (citizens), in interaction, thus serve as a cognitive filter, mediating the selection and
implications of the information shortcuts typical individuals rely upon in making political judgments.
In contrast to scholarship focused on longer-term public support for overseas conflicts, research
on the public’s immediate reactions to such events–the rally phenomenon–focuses far more on domestic
politics in general, and on statements by political elites in particular. In fact, the most widely accepted
domestic political explanation for the rally phenomenon holds that the extent of elite–and particularly
3
congressional (Hallin 1986, Althaus et al. 1996, Oneal et al. 1996, Bennett 1990, Zaller and Chiu
2000)–criticism of the president determines the magnitude of a post-use of force rally (Brody and
Shapiro 1989, Brody 1991). We refer to this as the “Opinion Indexing Hypothesis,” reflecting the
tendency of the public to “index” their opinions to the tenor of elite debate to which they are exposed.
A closely related “prevailing wisdom” in the literature holds that media coverage is itself
“indexed” to elite rhetoric in Washington (e.g., Hallin 1986, Bennett 1990, Entman and Page 1994, Zaller
and Chiu 2000, Bennett et al. 2006). We refer to this as the “Media Indexing Hypothesis.” The implication
is that the media are passive and non-strategic, faithfully reflecting the actual substance of elite debate.
Indeed, since, as Brody himself recognizes, citizens are exposed to elite debate primarily through the
media, the Opinion Indexing Hypothesis implicitly shares this assumption. Others go a step further,
arguing that elite debate actually bounds the range of arguments considered sufficiently “acceptable” to
receive any news coverage (Bennett 1990), or that support and consensus among elites will short-circuit
broader debate by constraining journalists’ willingness to challenge an administration (Hallin 1986).
In contrast, we argue that the nature and extent of media coverage of U.S. foreign policy crises is
driven less by political elites cowing journalists than by commonly held professional incentives and
norms that lead journalists to strongly prefer certain stories over others. For example, as Republican
Senator Hagel found when he characterized the Iraq war as similar to Vietnam, highlighting discord
within the president’s party is an especially attractive story, particularly when it occurs during unified
government. Conversely, there is relatively little incentive to cover boosterism of the president by his
own party, as George Allen discovered after his appearance on that same program.
Like event-based theories, the Opinion Indexing Hypothesis also assigns a passive role to
individual consumers. In contrast, consistent with substantial prior research, we argue that not all elite
statements are equally persuasive to different individuals. For example, opposition party endorsements
of or presidential party attacks on the president should be extremely credible to viewers because they are
4
atypical and represent costly signals (Dutton 1973, Eagly et al. 1978, Lupia and McCubbins 1998,
Groeling 2001). Similarly, typical individuals will likely view statements by their fellow partisan elites
as more credible than statements by opposition elites (Rahn 1993, Popkin 1994, Lupia and McCubbins
1998, Groeling 2001, Nelson and Garst 2005). In short, we argue that only by understanding the
individual incentives of, and strategic interactions between elites, the public, and the press can we
account for variations in public responses to presidential foreign policy initiatives.
Our theoretical framework draws on widely recognized characteristics of human information
processing, elite incentives and journalistic preferences. Taken individually, our assumptions are not
novel. However, we argue that combining these relatively common assumptions–concerning the distinct
preferences of the makers, transmitters, and receivers of news–yields a variety of non-obvious
predictions. The implications of our argument, in turn, extend well beyond foreign policy. Nonetheless,
we focus on foreign policy crises as a particularly interesting and useful application of our framework
because prior theories of public opinion and foreign policy have generally ignored the strategic incentives
of media actors and their potential effects on the nature of the information upon which distinct subgroups
of the public base their opinions. We also view foreign crises (particularly those involving military
mobilizations and conflicts) as an especially “hard case” in which to find an independent effect from
media or elite rhetoric because they involve life-and-death risks and large-scale movements of people and
equipment. Such crises thus tend to be unusually visible and salient to the public compared to the
material costs and benefits of most domestic policy initiatives, which tend to be observable only
gradually and primarily over the long term, if at all (Arnold 1992).
By analyzing network news coverage of congressional evaluations of the president and his
administration in periods surrounding all major U.S. uses of military force between 1979 and 2003, we
propose to demonstrate that even after controlling for a wide range of indicators of empirical “reality,”
communication still plays a crucial, independent role in influencing public support for the president
5
during foreign crises. We shall further show that, rather than simply parroting the statements of
Washington elites, public opinion in these crises varies systematically with the credibility of those
statements, as well as the institutional context in which political communication takes place and the
characteristics of the receivers; that is, depending on who the president is at the time of a crisis, who is
speaking about it, and who is listening to their rhetoric.
THEORETICAL FRAMEWORK AND HYPOTHESES
What Politicians Want from the Media: The most universally accepted assumption in U.S. electoral
politics is that politicians seek, first and foremost, re-election (Mayhew 1974). We generalize Mayhew’s
famous observation by assuming that politicians seek re-election both for themselves and their fellow
partisans. After all, winning a seat in the Congress holds dramatically different implications–both with
respect to resources available for subsequent election campaigns, and for a member’s ability to influence
public policy–if one is a member of the majority party (Cox and McCubbins 1993; Cox and Magar
1999). Winning election or majority party status, in turn, requires making one’s self and one’s fellow
partisans look good, while casting members of the opposition party in a negative light. The implication
for politicians’ preferences regarding media coverage is straightforward; typical politicians prefer stories
that praise themselves and fellow partisans, or criticize their opponents or the opposition party.
In the context of inter-branch relations, this further implies that–notwithstanding any journalistic
preferences for covering particular statements–members of the president’s party in Congress are likely
to offer rhetorical support for the president, while opposition party members should be more likely to
oppose him. While there are certainly periodic incentives for individual members to depart from these
strategies–particularly if they are running for president themselves or wish to gain additional press
coverage by taking “maverick” stances–the perceived novelty of such instances highlights the prevailing
baseline from which they depart.
If journalists do, as we shall demonstrate, consistently report discord among the president’s fellow
6
partisans more frequently than affirmation, there can be only two explanations. Either such coverage
must reflect journalists’ preferences, or elites from the president’s own party must be routinely
criticizing the president more often than they praise him during times of foreign crises–thereby
supporting the passive media assumption of the Media Indexing Hypothesis. We consider the latter
possibility highly improbable, especially given that in the most public of all representations–votes for or
against presidential initiatives in the legislature–recent presidents have typically received overwhelming
support from members of their own party and strong opposition from the opposing party.3
What Journalists Want from Politicians: Despite politicians’ best efforts to control their public
communication, journalists and news organizations maintain ultimate control over the content of their
news programs because of their function as “gatekeepers” of political news content. In deciding what
political material is or is not “news”, certain characteristics of stories or sources make them more (or
less) desirable for journalists. In particular, journalists generally prefer stories that are novel, conflictual,
balanced, and involve authoritative political actors (Graber 1997, Groeling 2001, PEJ 2002).
The most obvious characteristic of newsworthiness is that it entails a premium on stories that are
actually new. Informing readers or viewers of unexpected, inconsistent, novel, or surprising information
is the core value provided by news organizations. In fact, without novelty it makes very little sense to
speak of "news" organizations at all. This preference leads reporters to strongly resist attempts by
politicians to deliver "scripted," consistent messages to the public. As CBS's chief White House
correspondent noted when covering the 2004 Republican National Convention journalists want “to find
the inconsistency here, to find the people who aren't quite agreeing with the script that's going on any
given convention night, to get behind the story” (Kurtz 2004). In brief, a preference for novelty implies
simply that journalists place a premium on stories that are unusual. More formally, journalists prefer
stories that contain new or unexpected information to stories presenting old or expected information.
A second characteristic of "good" news is, ironically, a preference for bad news. Numerous
7
scholars (e.g., Sabato 1991, Patterson 1996, Cappella and Jamieson 1997) have observed that while
negativity and conflict have long been staples of American journalism, the news media have increasingly
embraced "attack journalism" and cynicism since the 1960s. Indeed, there seems to be consensus within
the scholarly literature that negativity is pervasive and dominant in modern news coverage.
While not all politicians go so far as former Vice President Spiro Agnew in characterizing the
media as "nattering nabobs of negativism," recent politicians appear to have shared the view that the
press favors negativity and conflict in their story choices. Early in his first year in office, President Bill
Clinton had already concluded that for the media, "success and lack of discord are not as noteworthy as
failure."4 As one journalist bluntly observed, “Well, journalists are always looking for conflict. That's
what we do.” (Saunders, in Kurtz 2004). Therefore, we argue that journalists prefer stories in which
political figures attack each other to stories in which political figures praise each other.
Considerable ink has been spilled debating whether the media might be more likely to attack
liberal or conservative points of view in their coverage. Tuchman (1972) famously argued that in part to
counter such bias accusations, journalists have a strong incentive to use procedures or strategic "rituals"
of objectivity in doing their jobs. The main ritual Tuchman and others discuss is presenting "both sides
of the story." News organizations, particularly broadcasters, have long followed this "balancing"
practice. For most of the 20th Century, broadcast stations and networks were held to an exceptionally
high standard of fairness through FCC regulation (the so-called “fairness doctrine”). Journalists have
also internalized these standards through professional ethics and norms, which require them to make
every effort “to assure that the news content is accurate, free from bias and in context, and that all sides
are presented fairly” (ASNE 2002). We thus assume journalists prefer stories that include both parties’
views to stories that only present the views of members of a single party.
Finally, journalists place a premium on getting the most authoritative and high-ranking possible
source. As Graber (1997, 116) argues, the "gatekeeping process winnows the group of newsworthy
8
people to a very small cadre of familiar and unfamiliar figures... predominantly political figures." Sigal
(1986) adds that "by convention, reporters choose authoritative sources over other potential sources" and
that "the higher up an official's position in government, the more authoritative a source he or she was
presumed to be, and the better his or her prospects for making the news" (20). Lippmann (1920)
concurred, arguing that: “The established leaders of any organization have great natural advantages.
They are believed to have better sources of information….It is, therefore, easier for them to secure
attention and speak in a convincing tone.” More formally stated, journalists prefer to include sources
with greater authority in their stories over less authoritative sources.
The top section of Table 1 applies these story characteristic preferences to four types of partisan
evaluations of the president.5 This allows us to determine which types of stories are most likely to gain
airtime. With respect to such evaluations, Table 1 shows that praise of the president by his own party–
which lacks novelty, balance, and conflict–is of little interest to journalists, especially during divided
government. In contrast, presidential party criticism of the president is highly attractive to journalists–
especially during unified government–because it is conflictual and novel.
[Table 1 here]
In contrast, evaluations of the president by the opposition party tend to be newsworthy
regardless of which party controls Congress, albeit somewhat more so in divided government. Such
comments are always either novel–if they support the president–or conflictual–if they criticize him.
Airing opposition party comments also adds balance to stories about the president and his policies.
Finally, journalists’ preference for authoritative sources leads to an over-representation of the majority
party in Congress. If the majority party happens to share the president’s party affiliation–i.e., in unified
government– this leads to the strongest possible incentive for journalists to air any intra-party criticism
of the president: Any such criticism is novel, conflictual, and authoritative. Our first hypothesis follows.
(H1) Oversampled Presidential Party Criticism: Because presidential party criticism is far more
9
newsworthy than presidential party praise, the news media will present more negative than
positive evaluations of the president by his own party in the news.
Salient Rally Events as Special Cases for Journalists: If the top section of Table 1 delineates the
newsworthiness of “politics as usual,” this raises the question of how newsworthiness during a major
foreign crisis might systematically differ. For much of the post-World War II era, the Republican and
Democratic parties are commonly viewed as having achieved near-consensus in foreign policy, especially
with respect to the Cold War. In explaining the shift of many in his party from prewar isolationism, Sen.
Arthur Vandenberg, the Republican chairman of the Senate Foreign Relations Committee, famously
explained, "Politics stops at the water's edge." Implicit in the very notion of a “rally-round-the-flag” is that
major international crises will induce each party to close ranks and increase its support for the president.
From a standpoint of newsworthiness, however, the impact is somewhat more complex. If
journalists expect partisans from both parties to rally behind the president when American troops are in
harm’s way, criticism of the president by either party should become even more newsworthy than
during non-crisis periods. The middle section of Table 1 illustrates this point. While this table tells us
little about each party’s intent to support the president in crisis periods, it does suggest that if any
members of either party choose to criticize the president, they should find journalists even more eager to
air their comments than during other times. Thus, we derive our second hypothesis:
(H2) Salient Rally Novelty: For MC’s from both parties, the criticism-to-praise ratio regarding the
president appearing in the news (that is, the amount of criticism, relative to praise) will be
greater during high-salience rally periods than during other periods.
What is Persuasive to Consumers? We now turn from the suppliers (the news media) to the
consumers of news (the audience). In determining each message’s effect on viewers, it is important to
consider not just the content of a message, but also its credibility. Parties do not “inject” messages into a
passive public; individuals process such messages and can accept or reject them depending in part on
10
their perceived credibility (Sniderman et al. 1991, Kuklinski and Hurley 1994, Druckman 2004).
One source of credibility for a message is the belief that the speaker and listener have common
interests (Crawford and Sobel 1982). Restated in partisan terms, this suggests that statements by a
listener’s own party will be regarded as more credible than those of the opposing party, all else equal.
We call this our Partisan Credibility Conjecture. This suggests a third hypothesis:
(H3) Partisan Credibility: Presidential evaluations by members of a given party will have a
stronger effect on that party’s identifiers' propensity to rally than will comments by
members of the other party.
Another important source of credibility derives from the interaction of source and message:
whether the message is costly to the speaker (Spence 1973). Typical individuals regard messages that are
harmful to the interests of the speaker as more credible than those that impose no costs (so-called “cheap
talk”).6 In the context of partisan messages, it follows that messages by partisan speakers that appear to
damage their own party or help the other party are regarded as more credible than messages that help
their own party or damage the other party. We term this our Costly Credibility Conjecture. Such costly
messages should be at least somewhat credible regardless of the party affiliation of the listener.
The bottom section of Table 1 summarizes the relative credibility of different partisan messages
about the president based on their partisan and costly credibility. It demonstrates the relatively weak
persuasive power of “politics as usual” statements (i.e., non-presidential party attacks on the president,
and presidential party praise). Such statements by members of the presidential (non-presidential) party
serve only to rally their own partisans, who in all likelihood already approved (disapproved) of the
president prior to any rally event, and hence cannot re-evaluate the president upward (downward)
following a use of military force (Baum 2002).
In contrast, non-presidential party praise should be exceptionally persuasive and beneficial to the
president, especially among non-presidential party members. If rally events produce bipartisan elite
11
support for the president, such support should be highly effective at moving public opinion–especially
among opposition identifiers–in support of the president. Similarly, if members of the president’s own
party attack him, the negative effects on public opinion should be dramatic, especially among the
president’s fellow partisans. In both cases, the media demand for such statements virtually ensures they
will receive coverage if offered, further magnifying their potential impact on opinion. Finally, because
independents lack a party affiliation and are therefore unaffected by partisan credibility, they should be
particularly influenced by the costly credibility of elite rhetoric. Two hypotheses follow:
(H4) Costly Credibility: Costly evaluations by members of a party will have a stronger effect on all
individuals' propensity to rally than will "cheap talk" evaluations by the same party.
(H5) Combined Credibility: Positive (negative) evaluations by non-presidential (presidential)
party elites, which have both costly and partisan credibility, will have the strongest effects on
presidential approval ratings for fellow members of their respective parties.
Credible Messages and Partisan Control of Government: Returning to the top of Table 1, we see
that the newsworthiness of the two most credible types of partisan evaluations noted in H5 (presidential
party criticism and non-presidential party praise) is systematically related to the partisan makeup of
government. Specifically, because the presidential (opposition) party is more (less) authoritative in
unified government, shifting from unified to divided government decreases the newsworthiness of the
most damaging type of message (presidential party criticism) while increasing the newsworthiness of
the most helpful messages (non-presidential party praise). This, in turn, seems likely to influence the
propensity of the public to rally under differing constellations of government. Two hypotheses follow.
(H6) Divided Rally Media Hypothesis: The proportion of credible praise to credible criticism in
the media will be greater in divided than in unified government.
(H7) Divided Rally Opinion Hypothesis: Aggregate public opinion rallies will be more positive
during divided than in unified government.
12
Salient Rally Events as Special Cases For Public Opinion: For the same reasons noted
previously for journalists, members of the public should be particularly attentive to such costly
evaluations in rally periods that involve U.S. casualties. Considerable research has shown that U.S.
casualties attract public attention and mediate public support for U.S. military conflicts (Mueller 1973,
Gartner and Segura 2000, Eichenberg 2005), while numerous studies (e.g., Hegre 2000, Oneal, et al.
2003, Gartzke and Gleditsch 2005) have employed the existence or absence of casualties in militarized
interstate disputes (MIDs) to distinguish relatively serious MIDs from less significant ones.
In addition, criticizing the president during a particularly high-profile foreign crisis is especially
risky. Research (e.g., Zaller 1994, Zaller and Chiu 2000) has shown that risk-averse members of Congress
typically prefer to avoid such criticism until the political ramifications of the crisis outcome are relatively
clear. While many scholars have argued that rising casualties depress public support for a conflict over
time (e.g., Mueller 1973, Gartner and Segura 2000), in the vast majority of cases we do not believe this
logic applies during the initial stages of a conflict (the rally period). Schwartz (1994), for instance, argues
that in the short-term, casualties usually harden the public’s resolve, consequently strengthening public
support for the use of force (see also Kull and Destler 1999).7 This raises the costly credibility of negative
comments by either party during rally periods involving U.S. casualties (in the short-run). Conversely, like
journalists, the public is likely to anticipate initial elite support for the president during salient rally periods
(e.g., those involving U.S. casualties). This reduces the costly credibility associated with supportive
comments by the opposition party, thereby mitigating their persuasive impact. For the presidential party,
however, comments supporting the president nearly always lack costly credibility. Thus, such comments
have little credibility to lose and should produce similar, limited effects on opinion both during and outside
of salient rally periods. Two final hypotheses follow:
(H8) Salient Rally Criticism: Negative evaluations by either party, which are most “costly” during
rally periods with U.S. casualties, will have a bigger effect on approval ratings than similar
13
comments during other periods.
(H9) Salient Rally Praise: During rally periods with U.S. casualties, positive evaluations by the nonpresidential party will produce smaller effects on approval ratings than similar comments
during other periods. Because they are cheap talk, positive presidential party comments should
be similarly unpersuasive in periods with and without casualties.
DATA AND METHODOLOGY
Mueller (1973) argues that for an event to be classified as a potential rally event, it should be
international, directly involve the president, and be “specific, dramatic and sharply focused” (1973:209).
Oneal et al. (1996:265) further restrict their definition of rally events to “major uses of force during a
crisis,” insuring that they are “considering only cases that were truly consequential for the U.S. and
salient to the public, necessary conditions for a rally…” Following Oneal et al., we restrict our analysis
to major uses of force during foreign policy crises. We employ an updated version of Baum’s (2002)
dataset, which, in turn, represents an update of Blechman and Kaplan’s (1978) dataset on political uses
of force (see also Oneal et al. 1996, Fordham and Sarver 2001). Again following Oneal et al. (1996), we
code all uses of force measuring levels 1-3 on Blechman and Kaplan’s (1978) scale as “major uses of
force.”8 Our data includes a total of 42 such events between 1979 and 2003 (hereafter "rally events").9
We collected data on all congressional comments on the president and the executive branch during
61-day windows surrounding each rally event, from 30 days before to 30 days after the announcement or
initiation of the major U.S. force deployment associated with each event. While we would prefer to have
gathered comprehensive measures of all sources of partisan rhetoric, the exceptional costliness of
this content analysis work–representing thousands of hours of research assistant labor over five
years–forced us to limit ourselves to the most important subset of these data (Hallin 1986, Bennett
1990, Zaller and Chiu 2000, Althaus et al. 1996, Oneal et al. 1996). While presidential rhetoric is, of
course, vital to the conduct of modern American politics (Kernell 1997), presidents tend to
14
uniformly support their own initiatives, leading to almost no variation in our key variables of
interest. For instance, in a multi-year content analysis of presidential rhetoric, Groeling (2001) finds
that over 90% of presidential self-evaluations are positive. In addition, as we explain below in our
analysis of partisan message credibility, such self-serving statements are cheap talk, and so should
generally be far less persuasive to typical voters than messages of support from across the aisle.
For each 61-day window, we first searched the Vanderbilt Television News Abstracts to locate
every appearance on the evening newscasts of ABC, CBS, and NBC by a senator or representative.10 Our
research assistants watched recordings or read verbatim transcripts of each selected story, coding the
statement's valence (positive, negative or neutral) along a number of issue dimensions (e.g. foreign policy,
budget, taxation), as well as the characteristics of the speaker (e.g., party, leadership status).11 (See online
data appendix for coding form.) All coded statements were direct quotes of an identifiable member of
Congress (hereafter “MC”) pertaining directly to the president. Each observation consists of a summary of
the content of a statement by a single MC in a single story. Although each statement might contain
multiple, distinct instances of praise or criticism of the president, we code all statements dichotomously on
both dimensions, separately recording the presence or absence of praise and/or criticism. 12
We assigned each story to two coders, working independently. Experienced graduate student
research assistants then reviewed and arbitrated any disagreements in the coding. Prior to arbitration,
inter-coder agreement on praise and criticism of the president was 95% and 88% for CBS and 86% and
96% for NBC, respectively.13 The arbitration process increases the reliability of our coding. In a random
sample of our data, our two graduate student arbitrators agreed on over 98% of all arbitration decisions,
producing a post-arbitration kappa score for our key causal variables of .86.14
We identified a total of 5302 pertinent congressional appearances on network evening news
programs during the 2115 days falling within ±30 day windows surrounding our 42 rally events.15 For our
public opinion analysis, we aggregate our data to the level of individual approval polls appearing within
15
our 61-day windows surrounding each event. This yields an average of 4.1 unique observations per event,
of which an average of 2.7 polls took place in the period after a major deployment was initiated or
announced. To mitigate serial autocorrelation we include the appropriate partisan presidential approval
poll lagged one period. This also accounts for the possibility that MC’s may base their decisions to
rhetorically oppose or support the president on their assessments of his ex ante political capital, or on
anticipated public reactions. We also transform our dependent variables into differences, for each partisan
subgroup, between approval at time t+1 and at time t. Finally, because the several observations associated
with each rally event are clearly not independent of one another, we cluster the standard errors by event.16
Control Variables: Many of our controls mirror those employed in previous studies of
presidential approval and the rally phenomenon (e.g., Oneal et al. 1996, Gartner and Segura 2000, Baum
2002, Nicholson, et al. 2002, Chapman and Reiter 2004). They are intended to account for the domestic
political circumstances surrounding each rally event, as well as the characteristics of the speaker
evaluating the president, of the adversary nation, of the rally event itself, and of the international
environment at the time of the event. For speaker characteristics, in addition to party affiliation (see
above), we include a dummy variable measuring whether MC’s are identified in a given story as leaders
of the House or Senate, their party, or a committee. For domestic political variables, we include the
number of mentions-per-poll-period of the adversary nation on the front page of the New York Times,
the monthly change in consumer sentiment (lagged one month),17 as well as dummies for presidential
and midterm election years, unified government, presidential transition periods, second term presidents,
and Democratic presidents. We also account for the number of days in between consecutive approval
polls and the number of appearances by MC’s on network evening newscasts during each poll period.
For adversary characteristics, we control for U.S. trade dependence and material capability ratio
vis-à-vis the adversary, and whether the adversary was a U.S. ally. For the international environment, we
include variables measuring the number of U.S. foreign policy crises in the year of a given event and
16
whether or not the event took place during the Cold War.18 Finally, for event characteristics, we include
dummies for whether an observation took place before or after the start dates of major U.S. force
deployments (or announcements of such), whether the U.S. goal was imposing “foreign policy restraint”,
“internal political change”, or “humanitarian intervention” (as defined by Jentleson 1992 and Jentleson
and Britton 1998), as well as whether the event was terrorism-related, involved a significant ground
invasion by U.S. troops,19 and lasted only one day. We also account for whether U.S. forces suffered any
combat deaths during a given poll period and the total number of U.S. casualties-per-poll-period (logged).
Finally, in order to increase our confidence that we have fully accounted for the unique
characteristics of each event, we polled 38 scholars with expertise in American foreign policy, asking
them to separately evaluate (on 0-10 scales) the extent to which, in their judgment, the events were
“successful” and “worthwhile” (based on their own cost-benefit assessment) for the United States.20 We
added the two items together to form a single “expert assessment” scale. We then regressed all of our
control variables on this summary indicator, and saved the residuals. The R2 was .72, indicating that our
control variables, excluding partisan rhetoric and lagged presidential approval, account for 72% of the
variance in our experts’ summary assessments of our 42 rally events. We employ the residual of our
experts’ summary assessments–that is, the exogenous portion–as a causal variable.21 (See Appendix for
more detailed descriptions of our control variables).
STATISTICAL RESULTS
Media Coverage Hypotheses: We begin with our Oversampled Presidential Party Criticism (H1),
and Salient Rally Novelty (H2) Hypotheses. Table 2 summarizes the valence of partisan evaluations in our
data. One noteworthy pattern is the overwhelming predominance of negative evaluations. Depending on
how we parse the data, between two thirds and four fifths of all evaluations featured on the network news
during our 61 day windows were negative. This pattern holds across networks, and also if we limit our
comparison to evaluations concerning only the president’s handling of foreign policy.22 Somewhat more
17
surprisingly, as Table 2 also indicates, the overwhelming predominance of negativity remains
largely unchanged during the period following the major U.S. deployments (or announcement of
such) surrounding our rally events, or during periods where the U.S. suffered casualties. In fact,
regardless of how we parse the data, criticism of the president and his administration actually
increases modestly following U.S. military deployments, with no statistically significant
differences during periods with American casualties. Table 2 also offers strong support for H1; no
matter how we slice the evaluations, a majority of all PP evaluations of the president are negative.23
One alternate explanation for this negativity may be the disproportionate weight our data places
on post-Cold War years, which accounts for a majority of our sample. Some scholars (e.g., Holsti 2004)
have conjectured that absent the unifying threat to national survival posed by the Soviet Union, domestic
politics may wield a stronger influence on American foreign policy in the post-Cold War era. Our results
offer at most limited support for this conjecture. While the middle section of Table 2 shows the level of
negativity in congressional evaluations by non-presidential party (henceforth “NPP”) MC’s did rise in
the post-Cold War period, that for presidential party (henceforth “PP”) MC’s actually declined
modestly, and both of these differences are small and statistically insignificant.
The middle section of Table 2 also tests our Salient Rally Novelty Hypothesis (H2). As predicted–
and contrary to conventional wisdom–due to their exceptional novelty, PP criticism in the news actually
increases by 12 percentage points (from 55 to 67% of all PP evaluations, p≤.01) following U.S. military
deployments (or announcements of such) during rally events. Conversely, PP praise declines by 11 points
(from 46 to 35%, p≤.05). Also as predicted, the onset of a major deployment or employment of force
yields a smaller and statistically insignificant 3 (2) percentage point increase (decrease) in NPP criticism
(praise). As one might expect, however, when we “raise the bar” and focus only on periods where it would
arguably be most politically risky for MC’s to criticize the president – which we operationalize as periods
in which the U.S. suffers casualties in early stages of major force deployments--we do in fact find,
18
inconsistent with H2, modest (albeit statistically insignificant) declines in criticism.
The far-right section of Table 2 provides strong support for H6 (Divided Rally Media
Hypothesis). Here, we break down the relative proportion of all partisan messages in unified and divided
government according to their partisan and costly credibility. As predicted by H6, the most damaging
PP evaluations (PP criticism) drop proportionately by over half from unified to divided government (33
vs. 16% of all evaluations in unified vs. divided government, p≤.001). Also consistent with H6, the most
credible form of praise of the president (NPP praise) nearly doubles, proportionately, in unified relative
to divided government (6 vs. 10% of all evaluations, p≤.02).24
In sum, Table 2 offers strong support for our media hypotheses, including clear support for the
Oversampled Presidential Party Criticism and Divided Rally Media Hypotheses, and qualified support
for the Salient Rally Novelty Hypothesis; supporting it for pre- vs. post-deployment periods, but
(unsurprisingly) less so for non-casualty vs. casualty periods.
Public Opinion Hypotheses: We turn next to our public opinion hypotheses, including the
Partisan (H3), Costly (H4) and Combined (H5) Credibility Hypotheses, the Divided Rally Opinion
Hypothesis (H7), as well as the Salient Rally Criticism (H8) and Praise (H9) Hypotheses. In order to
investigate the sensitivity of our results to model specification, we present three versions of our models.
The first excludes all controls, except for two measuring the number of days between each poll and the
number of MC evaluations-per-poll period. The second adds a series of domestic political controls,
while the third adds controls for characteristics of the event, adversary, and international environment.
The key causal variables measure the number of instances of praise or criticism of the president by either
party during a given poll period. Models 1-9 in Table 3 present the results from these tests.
[Table 3 here]
The first noteworthy pattern in Table 3 is the impressive consistency of the results on our key
causal variables across model specifications. With only one significant exception (PP criticism in the
19
political model) throwing the proverbial “kitchen sink” at our rhetoric indicators produces surprisingly
modest changes in their effects. While many of the controls are statistically significant, their effects appear
mostly orthogonal to our key causal variables. We can therefore proceed more confidently to interpreting
our results from the fully specified models. For ease of interpretation, we employ Clarify (King et al.
2000), a statistical simulation procedure, to calculate the expected values of our dependent variables as the
key causal variables vary by two standard deviations. This procedure also derives standard errors
surrounding the expected values, thereby allowing us to determine whether the differences in the effects of
the causal variables are themselves statistically significant. We present these results in Table 4.
[Table 4 here]
Beginning with party identifiers, Models 7 and 8 in Table 3 present the results from our fully
specified models investigating the effects of MC rhetoric on PP and NPP partisans, respectively. The top
and middle sections of Table 4 summarize the substantive effects of a two standard deviation increase in
each type of MC rhetoric. Among PP partisans, these results show strong support for the Partisan (H3),
Costly (H4), and Combined (H5) Credibility Hypotheses. Increased PP criticism, which should (per H5)
have the strongest persuasive impact for PP partisan, is associated with a large (3.1 percentage point)
and significant (p≤.05) decrease in approval. In contrast, the effect of NPP criticism, which lacks both
partisan and costly credibility for PP partisans, is small and insignificant. In addition, a two standard
deviation increase in NPP praise, which should (per H4) have a greater persuasive impact than “cheap
talk” NPP criticism, is associated with a relatively large (2.2 percentage point) and significant (p≤.05)
increase in PP approval. Consistent with H3, which would predict that shared partisanship should
convey some degree of credibility on otherwise “cheap” PP praise of the president, PP praise does have
a positive, albeit insignificant (somewhat inconsistent with H3) effect.
Turning to NPP partisans, consistent with Hypotheses 4-6, PP praise is not persuasive and is
associated with small and statistically insignificant effect. Consistent with H3, a two standard deviation
20
increase in NPP criticism yields a 5-point drop in NPP approval (p≤.10). The substantial (3.2 percentage
point) and highly significant (p≤.01) increase in NPP approval associated with NPP praise, in turn, also
provides strong support for Hypotheses 4-6.25 Presumably due to low partisan credibility, while correctly
signed and substantial in magnitude (-1.95), the effect of PP criticism fails to achieve statistical significance.
Also consistent with our hypotheses, the differences between the effects on approval ratings of highcredibility evaluation types (PP Criticism and NPP Praise) are themselves statistically significant for both PP
and NPP partisans (p≤.01), while those between low-credibility types (NPP criticism and PP praise) are not.
Finally, Model 9 in Table 3 presents the results from our fully specified model for Independents.
The substantive results shown in Table 4, in turn, offer strong support for H4 (Costly Credibility).
Moving from no MC evaluation to a two standard deviation increase in PP criticism yields about a 3percentage point decrease in approval (p≤.05). A comparable increase in PP praise, which lacks costly
credibility, yields a far smaller and insignificant effect. Costly NPP praise is associated with a 2.8-point
increase in approval (p≤.05), while “cheap talk” NPP criticism yields an insignificant (albeit substantial
in magnitude) 2.9-point drop in approval. Hence, consistent with H4, both high-credibility evaluation
types (PP criticism and NPP praise) produce significant effects, in the predicted directions, while both
low-credibility evaluation types (PP praise and NPP criticism) do not. Moreover, as with party
identifiers, the difference between the effects of high credibility evaluation types is itself significant
(p≤.01), while that between low-credibility types is far smaller and insignificant.
Next, we test H7 (Divided Rally Opinion Hypothesis) by using models 7, 8, and 9 to determine
the marginal predicted change in presidential approval, given the actual observed flows of partisan
rhetoric in both unified and divided government, across our 42 events. Consistent with H7, our results
predict that, on average, the actual partisan evaluations of the president in divided government would add
.80 approval points, while the flows in unified government would actually cost the president -.52
approval points. This 1.32 percentage point difference is significant at p≤.01. Even though our model
21
explicitly controls for divided government, these changes account for a relatively large proportion of the
average aggregate opinion changes across our 42 events (increases of .85 approval points overall, and .93
and .55 approval points during divided and unified government, respectively). This is consistent with
prior research (Meernik and Waterman 1996) that has found typical rallies –with the exception of major
wars (Chapman and Reiter 2004)–to be relatively small and ephemeral.26
We turn next to our Salient Rally Criticism (H8) and Praise (H9) Hypotheses which predict that
criticism by MC’s from either party during rally periods with U.S. casualties will have a bigger effect than
such evaluations in other periods, while positive evaluations from NPP MC’s (but not PP MCs) will have a
smaller effect than in other periods. To test these hypotheses we interact each type of rhetoric with a
dummy measuring whether or not the U.S. incurred casualties in a given poll period. The results for PP
partisans, NPP partisans and Independents are shown in Models 10, 11, and 12, respectively, of Table 3.
Because interaction terms are frequently associated with statistically significant effects even when the
variables themselves are insignificant, evaluating the substantive importance of such effects requires
assessing the significance of the differences in the dependent variable(s) produced by variations in the key
causal variables and interaction terms. Thus, the bottom section of Table 4 compares the substantive effects
of two-standard deviation increases in each type of rhetoric for non-casualty versus casualty periods.
Beginning with PP partisans, Table 4 shows, inconsistent with H8, that increased PP criticism
(which has high costly credibility) is associated with similarly large (-3.258 vs. -3.210) and significant
(p≤.05) effects during both non-casualty and casualty periods. However, somewhat more consistent with
H8, the same increase in NPP criticism is associated with larger and appropriately signed effects (.766
vs. -1.034) during casualty periods. Presumably due to low partisan credibility, however, neither type of
evaluation achieves significance. Consequently, this latter difference should be interpreted as at best
suggestive. Conversely, and more unambiguously consistent with H9, increased high costly credibility
NPP Praise yields considerably larger (2.349 vs. .766) and more significant (p≤.10 vs. insig.) effects
22
during non-casualty periods. Also consistent with H9, increases in PP praise, which nearly always lack
costly credibility, does not produce significant effects during either casualty or non-casualty periods.
Turning to NPP partisans, and consistent with H8, criticism by either party’s MC’s exerts a far
larger effect during casualty periods: 2.6 and 6 times greater for increased PP and NPP criticism,
respectively (-.999 vs. -2.644 and -5.230 vs. -29.535, respectively). Presumably due to differences in
partisan credibility, the effects of PP rhetoric are insignificant in both casualty and non-casualty periods,
while those of NPP rhetoric are significant in both cases (p≤.10 and p≤.05 for non-casualty and casualty
periods, respectively). Consistent with H9, in turn, increased PP praise is not associated with a significant
effect on NPP approval during either non-casualty or casualty periods. However, the corresponding increase
in NPP praise is associated with a far more significant, albeit somewhat smaller in magnitude, increase in
approval during non-casualty periods (3.438, p≤.01 vs. 4.137, insig.). Taken together, the results for PP and
NPP partisans offer fairly consistent, albeit imperfect support for H8 (Criticism) and H9 (Praise).
Finally, among Independents, the bottom section in Table 4 indicates that increased (high costly
credibility) PP criticism is associated with larger (-2.530 vs. -3.347) drops in approval (p≤.10 in both
instances) during casualty periods. The corresponding increase in high costly credibility NPP praise also
yields larger (.992 vs. 3.254) and more significant (insig. vs. p≤.01) increases in approval during noncasualty periods. Conversely, neither low costly credibility evaluation type (NPP criticism and PP praise)
achieves significance in either non-casualty or casualty periods. The latter pattern is consistent with H9.
However, the former is at least somewhat inconsistent with H8. Finally, though the effects of NPP
criticism are statistically insignificant, it is worth noting that, at least somewhat consistent with H8, they
are nearly 5 times larger in magnitude during casualty periods (-3.057 vs. -15.144). Overall, the data
support H8 and H9 unambiguously in 9 of 12 possible comparisons, and at least partially, in 10 (and
arguably 11) of 12 possible comparisons. Once again, this represents fairly strong support for our theory.
(See online appendix for summary of all hypothesis test results.)
23
A Few Words on Potential Counter-Arguments: We briefly address four potential criticisms,
including: (1) reverse causality; (2) that intrinsic characteristics of the events may drive both elite rhetoric
and public opinion; (3) MC criticism is intrinsically more significant than praise, and so journalists
“should” cover it and the public “should” value it more, and (4) differences in elite rhetoric in the news
“could” reflect the actual mix of elite rhetoric, rather than journalists’ preferences.
Beginning with reverse causality, we believe the concern that the changing patterns of evaluations
could reflect, rather than cause, changes in presidential popularity is unfounded for at least three reasons.
First, approval ratings at the time of the evaluation are directly factored in to our models through inclusion
of approval at time t as a lag term. Second, because we employ the difference between approval at time
t+1 and at time t as our dependent variable, a president’s unknown future approval logically cannot cause
present actions.27 Lastly, and most importantly, if one assumes that anticipated future increases in
presidential approval cause politicians to increase their support for the president, this should affect the
political calculations of both PP and NPP MC’s. Yet, in most cases we only observe significant effects for
praise from the NPP, while PP praise is insignificant in every case. Similarly, by this logic, there would be
no reason to expect that PP, but not NPP, criticism would be "caused" by anticipated future drops in PP
partisan approval, with NPP partisans responding only to criticism from their fellow partisan elites.
The second potential concern is that differences in the intrinsic characteristics of the events, rather
than in media coverage, may drive differences in MC rhetoric, and thus in public reactions. Yet our fully
specified model includes controls for a wide array of the unique characteristics of the events, including
whether the adversary was a U.S. ally, its military capabilities, U.S. trade relations with the adversary, the
U.S. “principal policy objectives”, the number of U.S. casualties, the number of U.S. foreign policy crises
under way at the time, whether the event involved a large-scale U.S. ground invasion or terrorism, whether
it took place during the Cold War, and whether it lasted one day. Moreover, wherever possible, we
gathered data based on the poll period, giving us an average of about four distinct observations per event.
24
This allows us to account for evolving circumstances as events unfold. Inclusion of our expert assessments
further enhances our confidence. The fact that our other controls explain well over 70% of the variance in
our experts’ summary assessment suggests that we have included a fairly comprehensive set of controls. In
the presence of all of these controls (including the exogenous portion of our expert assessments), it seems
improbable that some additional, unknown “unique” characteristics of the events are driving our results.
With respect to possible greater intrinsic value of critical evaluations, as also noted, our results
clearly show that NPP praise is strongly persuasive to all respondents save PP partisans. Similarly, if
negative evaluations were more intrinsically “important”, it seems likely that this would apply to all
critical statements by MCs of both parties, and not, as we find, just the subsets that are most credible to
their own partisans or independents.
Finally, we turn to the possibility that variations in actual elite rhetoric–rather than in journalists’
preferences–could be driving the differences we observe in elite rhetoric presented in the news. As noted
earlier, we believe it would be a truly heroic assumption to presume that, all else equal, elites prefer to
criticize their fellow partisan president far more than support him – which is the pattern we found in our
data. Still, because our dataset does not account for the complete universe of elite rhetoric offered to the
media, we cannot determine with certainty whether the observed patterns of coverage accurately reflect
the available population of potential evaluations.
While space limitations prohibit us from systematically addressing this concern here, elsewhere
(Baum and Groeling 2005) we address precisely this issue. In that study, to isolate the media’s
independent effect, we investigated a class of stories for which we can observe a full population of
potential elite rhetoric: all interviews with MC’s on NBC’s Meet the Press (MTP). Such interviews allow
elites to present their views in an unedited and comparatively unfiltered, “open mic” format, and are
routinely combed for fodder by all three networks’ evening news programs. While political interview
shows are not a perfect measure of the universe of elite rhetoric – after all, journalists select guests based
25
on presumptions of newsworthiness–examining which MC statements originating on these shows (that is,
the complete universe of available MC rhetoric on MTP) were actually selected for broadcast on the
evening news allows us far greater leverage to divine journalists’ preferences. Consistent with our theory,
we find that, relative to MTP, the evening news heavily over-represents PP criticism, while underrepresenting PP praise, especially during unified government. Consequently, we remain confident that
the rhetorical patterns we observed likely reflect the preferences of journalists more than the actual
population of statements offered by political elites (particularly elites in the presidential party).
CONCLUSION
The findings presented here hold potentially important implications for future leaders. While the
data appear to bear out Brody (1991) concerning the link between elite debate and the magnitude of
rallies, the process through which the content of this debate is selected and influences opinion is
substantially more complex and nuanced than previously assumed. Whether a given member of the
public rallies to support the president following the use of force is not simply a function of the overall
tenor of elite debate, but rather of (a) one's own partisan affiliation; (b) the partisan affiliations of the elite
debaters selected to appear in the media, (c) the credibility, or lack thereof, that results from the
interaction of these two factors, along with the costliness of the messages for the speakers, and (d) the
institutional incentives of journalists that lead them to cover or ignore particular speakers and messages.
We find little evidence that presidents can consistently expect to enjoy meaningful rallies when
they use force abroad, at least to the extent that rally magnitude does, as our evidence suggests, follow
from the nature and extent of elite debate presented in the media. Indeed, one of the most striking
patterns in our findings is the seemingly unyielding wave of negativity in media coverage of elite
discussion concerning the president and his policies. Most U.S. deployments of military force fail to
alter the unrelenting negative tone of elite discussion featured in the media.
Major conflicts may be a partial exception. When we limit our data to U.S. invasions involving
26
substantial incursions of ground forces (Grenada, Panama, Afghanistan and Iraq in 1991 and 2003), we
find somewhat less credible criticism and somewhat more credible praise, relative to the other events in
our data. During post-deployment periods surrounding ground invasions, highly credible NPP praise
nearly doubles proportionately (from 12 to 22% of all NPP evaluations), while (highly credible) PP
criticism falls from 69 to 42% of all PP evaluations. Applying the calculus employed earlier to predict
average rally size across our 42 events thus unsurprisingly yields larger predicted rallies during ground
invasions – an increase of about three approval points, on average, with, as before, larger rallies during
divided government. This is consistent with prior research (Chapman and Reiter 2004) that has found
evidence that substantial rallies are mostly limited to major wars and may help reconcile our finding of an
overwhelming overall negativity bias with the occasional emergence of substantial rally effects.
In addition to offering support for our theory concerning the effects of individual and
institutional factors in shaping the nature and extent of post-use of force rallies, our findings also hold an
important implication for diversionary war theory (Levy 1989). If presidents cannot be confident of
receiving favorable treatment in the media when they employ military force abroad – at least short of a
full-scale war like Operation Iraqi Freedom -- it seems highly unlikely that they would do so for purely
domestic political purposes. Our data suggest that attempting to divert public attention from domestic
difficulties through a use of force abroad is a highly risky strategy.
Prior to the midterm election of 2002, President George W. Bush crisscrossed the nation in what
was described as a "tireless campaign blitz" to win back Republican control of Congress. In relentlessly
campaigning for his fellow Republicans, Bush was gambling with the bipartisan prestige he had
accumulated following the September 11th attacks and the country's successful war in Afghanistan. But
perhaps more importantly for our story, Bush also worked to ensure that his rapidly approaching
confrontation with Iraq would take place under unified Republican control of government.
Ironically, our study suggests that while the midterm results may have made it easier for Bush to
27
win the congressional vote authorizing the war, the subsequent absence of credible praise from
authoritative Democratic sources made it far more difficult for him to later rally Democrats and
Independents in the electorate to his side. In addition, despite continuing support by the majority of
Republican elites, the news media’s elevation of highly-credible criticism from fellow Republicans such as
Hagel has helped push Bush’s approval ratings to historic lows. Viewed in this light, unified government
appears to be a mixed blessing, at least with respect to the president’s conduct of foreign affairs.
APPENDIX: CONTROL VARIABLES
Days Bet. Polls: Number of days between Gallup polls at time t and t+1.
MC Appearances: Number of appearances by MC’s on network newscasts during poll period.
Pres. Elec. Yr.: Coded 1 for cases occurring within 365 days of a presidential election, 0 otherwise.
Mid. Elec. Yr.: Coded 1 for observations occurring within 365 days of a midterm election.
Party Leader: Number of observations-per-poll-period in which MC evaluator was party leader.
Second Term. Coded 1 if a President is in second term in office.
∆Cons. Sent: Subtracts prior month's consumer sentiment score from the current month's score as
measured by the University of Michigan's Index of Consumer Sentiment.
Unified Gov’t: Coded 1 if presidential party has majority control of both chambers of Congress. (Control
is assumed to pass with the election of a new speaker or majority leader).
Transition: Coded 1 if observation occurs after election but prior to inauguration day.
Dem. President. Coded 1 if a Democrat was in office at the time of a given poll.
Any KIA. Coded 1 if the U.S. suffered any combat deaths during a given poll period.
Post-Deploy: Coded 1 if the statement was made on the day of the major U.S. force deployment, or
within 30 days after such an event.
Pre+Post Deploy: Coded 1 if statement was made both within 30 days after a force deployment and
within 30 days before another deployment.
28
lnKIA: Total number of U.S. casualties-per-poll-period (logged).
Major War. Coded 1 for U.S. invasions of Grenada, Panama, Iraq (1991 and 2003), and Afghanistan.
Post-Cold War: Coded 1 if observation occurred after fall of Berlin Wall (November 9, 1989).
NY Times Cov. Count of the number of mentions of the adversary nation on the front page of the New
York Times during a given poll period, divided by the average number of front page stories in the New
York Times during the same poll period.
One Day Event: Coded 1 if a given rally event lasted only one day.
Cap. Ratio: Correlates of War (COW) National Material Capabilities summary statistic (Singer & Small
1993). It takes the form of CA/(CA+CB), where CA = U.S. capabilities and CB = adversary capabilities.
Terrorism. Coded 1 if the event involved international terrorism.
US Crises-per-yr: Count of the number of foreign policy crises (Brecher and Wilkenfeld’s 2006) in the
same calendar year as a given event, in which the U.S. is the crisis actor. (This variable consistently
outperforms the number of rally events per year, drawn from the dataset employed in this study.)
US Ally: Coded 1 if the adversary is involved in a formal alliance relationship with the United States at the
time of a rally event, 0 otherwise. These data are derived from the Correlates of War Interstate Alliance
Data set, version 3.03 (Gibler and Sarkees, forthcoming).
Trade Depend.: This indicator is derived from the “UN World Trade Flows” data set (Feenstra et al.
2005). It represents the sum of U.S. exports to the adversary, as a proportion of all U.S. exports, plus
U.S. imports from the adversary, as a proportion of all U.S. imports.
Foreign Policy Restraint (FPR), Internal Political Change (IPC), Humanitarian Intervention (HI):
Coded 1 if a US goal in conflict was imposing FPR, IPC, or HI, respectively (Jentleson & Britton 1998).
Expert Assess.: Scale measuring extent to which, on average, 38 foreign policy experts considered each
event “successful” and “worthwhile” for the U.S. The scale runs from -5 to +5, with -5 (+5)
indicating least (most) successful or worthwhile (µ = 5.57; σ = 3.67).
29
REFERENCES
Althaus, Scott L., Jill A. Edy, Robert M. Entman, and Patricia Phalen. 1996. Revising the Indexing
Hypothesis: Officials, Media, and the Libya Crisis. Political Communication 13: 407-421.
Almond, Gabriel A. 1950. The American People and Foreign Policy. New York: Praeger.
American Society of Newspaper Editors. 2002. ASNE Statement of Principles. Revised Aug. 28, 2002.
Arnold, R. Douglas. 1992. The Logic of Congressional Action. New Haven, CT: Yale University Press.
Baum, Matthew. A. 2002. The Constituent Foundations of the Rally-Round-the-Flag Phenomenon.
International Studies Quarterly 46:263-298
_____. 2003. Soft News Goes to War. Princeton NJ: Princeton University Press.
_____ and Tim Groeling. 2005. Resetting the Index: Media Bias and Elite Foreign Policy Evaluations,
1979-2003. Presented at 2005 Meeting of the American Political Science Association, Washington D.C.
Bennett, W. Lance, Regina G. Lawrence, and Steven Livingston. 2006. None Dare Call it Torture:
Indexing and the Limits of Press Independence in the Abu Ghraib Scandal. Journal of Communication
56(3): 467-485.
Bennett, W. Lance. 1990. Toward a Theory of Press-State Relations in the United States. Journal of
Communication. 40:2.
Berinsky, Adam. J. 2005. Assuming the Costs of War: Events, Elites, and American Public Support for
Military Conflict. Unpublished Manuscript, Massachusetts Institute of Technology.
Blechman, Barry M. and Stephen S. Kaplan. 1978. Force Without War: U.S. Armed Forces as a Political
Instrument. Washington D.C.: Brookings Institution.
Brody, Richard. 1991. Assessing Presidential Character. Stanford: Stanford University Press.
_____ and Catherine R. Shapiro. 1989. Policy failure and public support: The Iran-Contra affair and
public assessment of President Reagan. Political Behavior 11:353-370.
Cappella, Joseph N. and Kathleen H. Jamieson. 1997. Spiral of Cynicism: The Press and the Public
30
Good. New York: Oxford University Press.
Chapman, Terrence, L. and Dan Reiter. 2004. The United Nations Security Council and the Rally
‘Round the Flag Effect. Journal of Conflict Resolution 48: 886-909.
Congressional Quarterly Almanac. 1953-2000. Washington, D.C.: CQ Press.
Cox, Gary and Mathew McCubbins. 1993. Legislative Leviathan: Party Government in the House.
Berkeley: University of California Press.
Cox, Gary and Eric Magar. 1999. How Much is Majority Status in the U.S. Congress Worth? American
Political Science Review. 93: 299-309.
Crawford, Vincent and Joel Sobel. 1982. Strategic Information Transmission. Econometrica. 50:1431-51.
Druckman, James N. 2004. Political Preference Formation: Competition, Deliberation, and the
(Ir)relevance of Framing Effects. American Political Science Review 98(4): 671-86.
_______. 1999. The Limits of Political Manipulation: Psychological and Strategic Determinants of
Framing. Ph.D. Dissertation, University of California, San Diego.
Dutton, Donald. 1973. The Maverick Effect: Increased Communicator Credibility as a Result of
Abandoning a Career. Canadian Journal of Behavioral Science 5:145-151.
Eagly, Alice, Wendy Wood and Shelly Chaiken. 1978. Causal Inferences about Communicators, and
their Effect on Opinion Change. Journal of Personality and Social Psychology 36:424-435.
Eichenberg, Richard. 2005. Victory has Many Friends: U.S. Public Opinion and the Use of Military
Force, 1981-2005. International Security. 30: 140-177.
Entman, Robert M. 2004. Projections of Power. Chicago: University of Chicago Press.
_________ and Benjamin Page. 1994. The News Before the Storm. In Bennett, W. Lance and
David L. Paletz, editors. Taken by Storm: The Media, Public Opinion, and U.S. Foreign Policy
in the Gulf War, pp. 82-101. Chicago: University of Chicago Press.
Feaver, Peter D. and Christopher Gelpi . 2004. Choosing Your Battles: American Civil-Military Relations
31
and the Use of Force. Princeton, N.J.: Princeton University Press.
Feenstra, Robert C., Robert E. Lipsey, Haiyan Deng, Alyson C. Ma, and Hengong Mo. 2005. World
Trade Flows: 1962-2000. Working Paper 11040. Cambridge, MA: National Bureau of Economic
Research. Available at: http://www.nber.org/papers/w11040.
Fordham, Benjamin, and Christopher Sarver. 2001. Militarized interstate disputes and United States uses
of force. International Studies Quarterly 45:455-466.
Gartner, Scott Sigmund and Gary M. Segura. 2000. Race, Casualties, and Opinion in the Vietnam War.
The Journal of Politics 62: 115-146.
Gartzke, Erik A. and Kristian S. Gleditsch. 2005. Identity and Conflict: Ties that Bind and Differences
that Divide. European Journal of International Relations 12:1: 53-87
Gibler, Douglas M., and Meredith Sarkees. Forthcoming. Measuring Alliances: the Correlates of War
Formal Interstate Alliance Data set, 1816-2000. Journal of Peace Research.
Graber, Doris. 1997. Mass Media and American Politics. Washington, D.C.: CQ Press.
Groeling, Tim. 2001. When Politicians Attack: The Causes, Contours, and Consequences of Partisan
Political Communication. Ph.D. Dissertation, University of California, San Diego.
_____ and Samuel Kernell. 1998. Is Network News Coverage of the President Biased? Journal of
Politics 60: 1063-87.
Hallin, Daniel. C. 1986. The 'Uncensored War'. The media and Vietnam. Berkeley: UC Press.
Hegre, Håvard, 2000. Development and the Liberal Peace: What Does it Take to Be a Trading State?
Journal of Peace Research 37: 5--30.
Herrmann, Richard, James Voss, Tonya Schooler, and Joseph Ciarrochi. 1997. Images in International
Relations: An Experimental Test of Cognitive Schema. International Studies Quarterly 41: 403-433.
Herrmann, Richard, Philip Tetlock, and Penny Visser. 1999. Mass Public Decisions to Go to War: A
Cognitive-Interactionist Framework. American Political Science Review 93: 553-573
32
Holsti, Ole. 2004. Public Opinion and American Foreign Policy. Ann Arbor, MI: Univ. of Michigan Press.
Hurwitz, Jon and Mark Peffley. 1987. How Are Foreign Policy Attitudes Structured? A Hierarchical
Model. American Political Science Review 81: 1099-1120.
Jentleson, Bruce. W. 1992. The pretty prudent public: post post-Vietnam American opinion on the use of
military force. International Studies Quarterly 36:49-74.
_____ and Rebecca L. Britton. 1998. Still Pretty Prudent: Post-Cold War American Public Opinion on the
Use of Military Force. Journal of Conflict Resolution 42: 395-417.
Kernell, Samuel. 1997. Going Public: New Strategies of Presidential Leadership. 3rd ed. Washington,
D.C.: CQ Press.
King, Gary., Michael Tomz, and Jason Wittenberg. 2000. Making the Most of Statistical Analyses:
Improving Interpretation and Presentation. American Journal of Political Science 44:341-355.
Koeske, G. E, and Crano, W. D. 1968. The effect of Congruous and Incongruous Source-Statement
Combinations Upon the Judged Credibility of a Communication. Journal of Experimental Social
Psychology 4:384-399.
Kuklinski, James and Norman Hurley. 1994. On Hearing and Interpreting Political Messages: A
Cautionary Tale of Citizen Cue-Taking. Journal of Politics 56:729-751.
Kull, Steven and Clay Ramsey. 2001. The Myth of the Reactive Public: American Public Attitudes on
Military Fatalities in the Post-Cold War Period. In Phillip Everts and Pierangelo Isneria, (Eds.). Public
Opinion and the International Use of Force. London: Routledge.
Kull, Steven and I.M. Destler. 1999. Misreading the Public: The Myth of a New Isolationism.
Washington D.C.: Brookings.
Kurtz, Howard 2004. Republican Convention Gets Under Way; Kerry Interviewed on 'The Daily Show'
CNN’s Reliable Sources. 29 August.
Levy, Jack S. 1989. The Diversionary Theory of War: A Critique. In Handbook of War Studies, edited by
33
Midlarsky, M.I. New York: Unwin-Hyman.
Lippmann, Walter. 1920. Liberty and the News. New York: Harcourt, Brace, and Howe.
_____ 1934. The Method of Freedom. New York: Macmillan.
Lupia, Arthur and Mathew D. McCubbins 1998. The Democratic Dilemma: Can Citizens Learn What
They Need to Know? Cambridge: Cambridge University Press.
Mayhew, David. 1974. Congress: The Electoral Connection. New Haven: Yale University Press.
Meernik, James and Peter Waterman. 1996. The Myth of the Diversionary Use of Force by American
Presidents. Political Research Quarterly 49 (3): 573-590.
Mueller, John E. 1973. War, Presidents and Public Opinion. New York: John Wiley & Sons, Inc.
Nelson, Thomas and Jennifer Garst. 2005. Values-based Political Messages and Persuasion:
Relationships among Speaker, Recipient, and Evoked Values. Political Psychology 26:489-516.
Nicholson, S.P., G.M. Segura, and N.D. Woods. 2002. Presidential Approval and the Mixed Blessing of
Divided Government. The Journal of Politics 64(4): 701-720.
Oneal, John, Bruce Russett, and Michael Berbaum. 2003. Causes of Peace: Democracy, Interdependence,
and International Organizations, 1885-1992. International Studies Quarterly 47: 371-93.
Oneal, John R., Brad Lian and James H. Joyner, Jr. 1996. Are the American People Pretty Prudent?
Public Responses to U.S. Uses of Force, 1950-1988. International Studies Quarterly 40:261-280.
Page, Benjamin I., and Marshall M. Bouton. 2006. The Foreign Policy Dis*Connect: What
Americans WANT from Our Leaders but DON’T Get. Chicago: University of Chicago Press.
Patterson, Thomas. 1996. Bad News, Period. PS: Political Science and Politics 29:17-20.
Perla, Hector, Jr. 2005. Days of Decision: A Framing Theory of Public Opposition to the Use of Force
Abroad. Presented at annual meeting of the American Political Science Association, Washington D.C.
Popkin, Samuel. 1994. The Reasoning Voter (Second Edition). New York: Univ. of Chicago Press.
Project for Excellence in Journalism (PEJ). 2002. Local TV News Project-2002.
34
Rahn, Wendy. 1993. The Role of Partisan Stereotypes in Information Processing about Political
Candidates. American Journal of Political Science 37: 472-496.
Rosenau, James N. 1961. Public Opinion and Foreign Policy. New York: Random House.
Sabato, Larry. 1991. Feeding Frenz. New York: Free Press.
Schwartz, Benjamin C. 1994. Casualties, Public Opinion, and U.S. Military Intervention:
Implications for U.S. Regional Deterrence Strategies. Santa Monica, CA: Rand.
Sigal, Leon V. 1986. Sources Make the News, in Reading the News: A Pantheon Guide to Popular
Culture, Manoff, R.K. and M. Shudson, eds. New York: Pantheon Books.
Singer, J. David, and Melvin Small. National Material Capabilities Data, 1816-1985 [Computer file].
Ann Arbor, MI: J. David Singer, Univ. of Michigan, and Detroit, MI: Melvin Small, Wayne State
Univ. [producers], 1990. Ann Arbor, MI: ICPSR [distributor], 1993.
Sniderman, Paul, Richard Brody, and Philip Tetlock. 1991. Reasoning and choice: explorations in
political psychology. New York: Cambridge University Press.
Spence, A. Michael. 1973. Market Signaling. Cambridge, MA: Harvard University Press.
Tuchman, Gaye. 1972. Making News: A Study in the Construction of Reality. New York: MacMillan.
Walster, Elaine, Elliot Aronson and Darcy Abrahams. 1966. On Increasing the Persuasiveness of a Low
Prestige Communicator. Journal of Experimental Social Psychology 2: 325-42.
Zaller, John. 1994. Elite leadership of Mass Opinion: New evidence from the Gulf War. In Bennett, W.
Lance and David L. Paletz, editors. Taken by Storm: The Media, Public Opinion, and U.S.
Foreign Policy in the Gulf War, pp. 82-101. Chicago: University of Chicago Press.
Zaller, John, and Dennis Chiu. 2000. Government’s Little Helper: U.S. Press Coverage of Foreign Policy
Crises, 1946-1999. In Nacos, Brigitte L., Robert Y. Shapiro, and Pierangelo Isernia, editors,
Decisionmaking in a Glass House. New York: Rowman & Littlefield.
35
TABLE 1. Newsworthiness, Novelty, and Credibility of Rhetoric Regarding President
by Elites from Presidential Party (PP) and Non-Presidential Party (NPP)
PP Praise PP Criticism NPP Praise NPP Criticism
A. Newsworthiness of Partisan Evaluations of the President
Novelty
Low
High
High
Low
Conflict
Low
High
Low
High
Balance
Low
Low
High
High
Authority (UG)
High
High
Low
Low
Authority (DG)
Low
Low
High
High
B. Change in Novelty During Salient Rally Periods
Novelty During Salient Rallies
Low
Higher
Lower
Higher
C. Partisan and Costly Credibility, by Party of Speaker and Viewer
Costly Credibility
All Partisans and Independents
Low
High
High
Low
Presidential Partisans
High
High
Low
Low
Independents
Low
Low
Low
Low
Non-Presidential Partisans
Low
Low
High
High
Partisan Credibility
36
\TABLE 2: Summary of Valence in Congressional Evaluations of President (Percent of all MC Messages, by Type)
All
FP
Pres.
Non-Pres
Costly
Partisan Credibility
Issues
Only
Party
Party
Credibility
High
Low
ABC
Percent Negative
N=361
80
N=148 Cold War
73
Percent Negative
N=130
64
N=207
86
Percent Positive
CBS
Percent Negative
Percent Positive
NBC
Percent Negative
Percent Positive
21
N=384
77
26
N=459
80
21
27
Percent Positive
N=143 Post-Cold War
62
Percent Negative
41
Percent Positive
N=188 Unified Government
73
Percent Negative
28
Percent Positive
39
N=262
61
40
N=150
69
32
15
N=605
88
13
N=163
90
12
Overall
Percent Negative
Percent Positive
N=1204
79
22
N=479 Divided Government N=242
70
Percent Negative
58
31
Percent Positive
45
N=649
87
14
Pre-Deployment
Percent Negative
Percent Positive
N=560
77
24
N=238 Pre-Deployment
67
Percent Negative
34
Percent Positive
N=171
55
46
N=389
86
15
Post-Deployment
Percent Negative
Percent Positive
N=644
81
21
N=241 Post-Deployment
72
Percent Negative
29
Percent Positive
N=221
67
35
N=423
89
13
No Casualties
Percent Negative
Percent Positive
N=1079
80
22
N=389 No Casualties
70
Percent Negative
31
Percent Positive
N=341
62
40
N=738
88
13
PP viewers
Unified Government (n=313)
High
(PP-) 33
(NPP+) 6
Low
(PP+) 15
(NPP-) 47
Divided Government, (n=891)
High
(PP-) 16
(NPP+) 10
Low
(PP+) 12
(NPP-) 63
NPP viewers
Unified Government (n=313)
High
(NPP+) 6
(PP-) 33
Low
(NPP-) 47
(PP+) 15
Divided Government, (n=891)
High
(NPP+) 10
(PP-) 16
Low
(NPP-) 63
(PP+) 12
Any Casualties
N=125
N=90 Any Casualties
N=51
N=74
Percent Negative
74
68
Percent Negative
61
82
Percent Positive
27
32
Percent Positive
39
19
Notes: (1) Sums exceed 100% because some evaluations include both praise and criticism.; (2) ABC=ABC World News Tonight,
CBS=CBS Evening News, and NBC=NBC Nightly News; (3) PP+ = praise by presidential party MC’s, PP- = criticism by presidential
party MC’s, NPP+ = praise by non-presidential party MC’s, and NPP- = criticism by non-presidential party MC’s.
37
38
39
TABLE 4. Effect of Different Types of Rhetoric on Presidential Approval
Marginal Effect Difference from
of Evaluation*
No Evaluation
Presidential Party Approval (No evaluation=0.225)
PP Praise
1.117
0.892
PP Criticism
-2.875
-3.100*
NPP Praise
2.388
2.163*
NPP Criticism
0.916
0.691
Non-Presidential Party Approval (No evaluation=2.847)
PP Praise
1.288
-1.559
PP Criticism
0.901
-1.946
NPP Praise
6.081
3.234**
NPP Criticism
-2.103
-4.950^
Independents Approval (No evaluation=1.889)
PP Praise
1.187
-0.702
PP Criticism
-1.147
-3.036*
NPP Praise
4.733
2.844*
NPP Criticism
-1.054
-2.943
Casualty (KIA) vs. Non-Casualty (No KIA) Period Interaction Models
No KIA
KIA
Marginal
Difference
Marginal
Difference
Effect of
from No
Effect of
from No
Evaluation
Evaluation
Evaluation Evaluation
Presidential Party Approval
(No evaluation or KIA = -4.483; No evaluation with KIA = 8.315)
PP Praise
-3.020
1.463
9.081
0.766
PP Criticism
-7.741
-3.258*
5.105
-3.210*
NPP Praise
-2.134
2.349^
8.911
0.596
NPP Criticism
-3.717
0.766
7.281
-1.034
Non-Presidential Party Approval
(No evaluation or KIA = -0.508; No evaluation with KIA = 13.048)
PP Praise
-2.554
-2.046
18.220
5.172
PP Criticism
-1.507
-0.999
10.404
-2.644
NPP Praise
2.930
3.438**
17.185
4.137
NPP Criticism
-5.738
-5.230^
-16.487
-29.535*
Independents Approval
(No evaluation or KIA = -2.256; No evaluation with KIA = 13.210)
PP Praise
-2.539
-0.283
15.506
2.296
PP Criticism
-4.786
-2.53^
9.863
-3.347^
NPP Praise
0.998
3.254**
14.202
0.992
NPP Criticism
-5.313
-3.057
-1.934
-15.144
*based on two standard deviation increase in type of rhetoric, with other types
of rhetoric held constant at zero
40
NOTES
1
Our search of Lexis-Nexis’ online transcripts produced 9 hits for stories that only mentioned Allen,
and 277 that only mentioned Hagel (61 stories mentioned both).
2
Individuals also employ other heuristics in evaluating foreign policy, such as accessible “images” of
potential adversaries (e.g., enemy vs. friend) and core values, such as isolationism vs. internationalism
(Herrmann et al. 1997, Holsti 2004). Still, elite communication plays an important role in priming such
images and values, and thereby framing events for individuals. Some research (Herrmann et al 1999)
has found that party ID is not a good predictor of public support for military conflict. However, party
does mediate elites’ capacity to successfully frame events for different individuals (Druckman 2004).
3
Congressional Quarterly’s reports that since the Eisenhower Administration, an average of about two-
thirds of presidentss’ fellow partisans support them on votes where they stake a position, with presidents
since Reagan greatly exceeding that average. Conversely, opposition party support for presidents is
generally low, with no president managing to break even on such votes (CQ Almanac 1953-2000).
4
From a May 7, 1993 Clinton press conference, in response to a reporter’s question as to why his job
approval had dropped by 15 points in two months.
5
Because these evaluations are all directed at the president or administration, the stories already
implicitly contain some exposition of the president’s or administration’s position.
6
Related lines of inquiry is research in social psychological into the influence of “incongruous”
(Walster et al. 1966, Koeske and Crano 1968) or “disconfirming” messages (Eagly et al. 1978).
7
However, we nonetheless seek to isolate the salience component of the effects of casualties in our
statistical models by separately controlling for expert assessments of whether each U.S. use of force
was “successful” and “worthwhile.”
8
Following Baum (2002), we exclude several events inconsistent with these definitions, such as long-
41
scheduled military exercises, cancellation of a previously scheduled force withdrawals, or events that
clearly were not major uses of force during a U.S. foreign policy crisis (e.g., U.S. support for withdrawal
of U.N. forces from Somalia in January-March 1995, long after the U.S. withdrew its forces).
9
Our complete list of rally events may be found in an online data appendix at http://xxx.
10
Videotapes were acquired from Vanderbilt's Television News Archive and from UCLA's News and
Public Affairs video archive. Transcripts were accessed on Lexis-Nexis, where available.
11
Coders were UCLA and UCSD undergraduates. Before coding, they attended an orientation to the
coding scheme with one of the principal investigators or their two graduate research assistants, and then
practiced using a series of five online interactive practice-coding sessions.
12
Any additional utility from coding each individual critique within a member statement would be
outweighed by the exponential increase in complexity for our coding scheme. Our Count variable also
accounts for news appearances by MC’s during that did not include a codable evaluation.
13
Pre-arbitration kappa scores for these variables were .44 and .51, respectively, for CBS, and .52 and .48,
respectively, for NBC. Altman (1991: 404) characterizes this as "moderate" agreement. Our inter-coder
agreement for ABC was 80%. (Due to differences in coding procedures, Kappa is unavailable for ABC.)
14
While the coding form has remained constant, we implemented some improvements in the coding
process over time. For example, for a subset of ABC data, students hand-coded the stories, met to
compare their coding, and submitted their consensus results to a graduate student RA for further
examination. All of the NBC and CBS data, and the remainder of the ABC data, were submitted online,
with students unaware of the identity of their coding partner. We excluded a small subset of observations
in which tapes or transcripts were damaged or unavailable.
15
About 8.6% of our coded evaluations (457 out of 5302) occur fewer than 30 days before one rally and
fewer than 30 days after another rally. In all cases where sequence matters in our analysis, we count any
42
overlapping days as "after" the prior event, rather than "before" the next event.
16
We tested our models with event-specific fixed effects. The results were in many respects
comparable to those with clustered errors. Given our limited number of observations, however, we
have insufficient statistical leverage to be confident in the reliability of a fixed effects specification.
17
Changes in consumer sentiment outperformed a variety of other macro-economic indicators.
18
A post-9/11 dummy proved insignificant and had no material effect on our results.
19
Events meeting this definition include Grenada, Panama, Afghanistan, and Iraq (1991 and 2003).
20
We contacted 96 foreign policy experts via email. Our response rate was 40%.
21
The summary expert assessments for each event may be found in our online appendix.
22
Aggregating per rally event, we find 7.14 MC criticisms of the president (1.57 and 5.57 critical
comments by PP and NPP MC’s, with standard deviations of 1.90 and 5.57, respectively),
compared to just 2.25 supportive comments (1.26 and .79 by the PP and NPP, with standard
deviations of 1.26 and .79, respectively).
23
Unfortunately, we cannot, within the confines of our data, definitively prove that this dramatically
skewed distribution results from journalists’ choices, rather than a conscious choice by PP partisans to
attack their leader nearly twice as often as they praise him in the news. However, if one accepts what we
consider an extremely modest assumption–that PP partisans do not typically attack their fellow partisan
president far more than support him–then our empirical results clearly support the hypothesis. Moreover,
even if we exclude the one noteworthy episode in our data where PP partisan attacks on their own
president are likely to have been relatively common – during the 1998 Lewinsky scandal – the overall
pattern changes hardly at all. Elsewhere, we confront this “unobserved population” problem directly
(Groeling and Kernell 1998, Baum and Groeling 2005), and find that the news media do, in fact, oversample criticism, particularly from the president’s party (see discussion in “counter-arguments” section).
43
24
Baum (2003) reports evidence of a trend, between 1953 and 1998, toward larger rallies among the
least educated Americans, but not among their highly educated counterparts. At first glance, this seems
inconsistent with our findings of (a) overwhelmingly critical elite rhetoric during rally events since 1979,
and (b) a strong relationship between such coverage and rally magnitude. However, education levels in
America have risen such that the least-educated group has constricted as a proportion of the public since
the 1950s. It may simply be the case that smaller rallies among highly educated Americans outweigh the
effects of larger rallies among the least-educated citizens, who might, in turn, be less able to reliably
distinguish between credible and non-credible praise and criticism. Also, Baum’s time series extends far
longer than our data, making it difficult to draw direct comparisons between the two studies.
25
While the magnitude of the difference is actually largest for NPP criticism in the fully specified
model, the base and political models show larger, more significant differences for NPP praise.
26
Our marginal aggregate opinion change for each event represents an equally weighted average
of the predicted change for independents, presidential, and non-presidential party members. The
average observed change in approval across all of our events was an increase of .846 for presidential
party members, 1.06 for members of the non-presidential party, and .87 for independents. If we include
all controls, set at their mean values (except the divided government dummy), the model predicts
somewhat larger (smaller) rallies in divided (unified) government.
27
Of course, in some cases future presidential approval can be accurately forecast. In the case of rally
events such as the 9/11 attacks, it was probably clear to most politicians that the public would rally
around George W. Bush. But in most cases, it seems unlikely an MC could predict a president's future
popularity with enough certainty to affect his or her present actions.
Online Supplemental Appendix
SUMMARY OF HYPOTHESES, TEST RESULTS, AND LOCATIONS
Proportion
(%) of Tests
Supporting
Hypothesis
Where Tested?
Hypothesis
H1: Oversampled Pres. Party Criticism
Table 2
1/1 (100%)
H2: Salient Rally Novelty
H3: Partisan Credibility
Table 2
1/2 (50%)a
Models 7&8 of Table 3;
3/4 (75%)
Simulations in Table 4
H4: Costly Credibility
Models 7, 8 &9 of Table 3;
5/6 (83%)
Simulations in Table 4
H5: Combined Credibility
Models 7&8 of Table 3;
2/2 (100%)
Simulations in Table 4
H6: Divided Rally Media Hypothesis
H7: Divided Rally Opinion Hypothesis
H8: Salient Rally Criticism
Table 2
1/1 (100%)
Models 7, 8 &9 of Table 3
1/1 (100%)
Models 10, 11, and 12 of Table
4/6 (67%)b
3, Simulations in Table 4
H9: Salient Rally Praise
Models 10, 11, and 12 of Table
5/6 (83%)b
3, Simulations in Table 4
TOTALS:
a
23/29 (~80%)
This represents a conservative criterion for “support” of H2. The unsupportive result
represents a case where the theory would arguably predict a negative result. Hence, this
case “could” be interpreted as supportive.
b
These represent conservative estimates of the proportions of tests supporting H8 and H9. If
one counts results consistent with the hypotheses in relative magnitudes and valences, but
which fail to achieve standard levels of statistical significance, the overall proportion of
supportive results for H8 and H9 increases to 83% or 92% (from the above-reported 75%),
depending on the stringency of the criterion for counting a result as “supportive.”
EVENT LIST
1. Hostage crisis in Iran, November 1979.
2. Soviet invasion of Afghanistan: Carter Doctrine, January 1980.
3. Marine Barracks Bombing, October 1983.
4. Invasion of Grenada, October 1983.
5. Further attacks on/by U.S. troops in Lebanon. December 1980.
6. Operation El Dorado Canyon: U.S. airstrikes against Libya in response to Berlin disco
bombing. April 1986.
7. Operation Prairie Fire: U.S. engages Libyan aircraft, ships and missile sites around Gulf
of Sidra. April 1986.
8. U.S.S. Stark attacked by a missile. May 1987.
9. U.S.S. Vincennes shoots down Iranian civilian airliner. July 1988.
10. Response to Pan Am Flight 103 destruction. December 1988.
11. Two carriers, battleship groups moved to eastern Mediterranean, Persian Gulf, Arabian
Sea after killing of Col. William Higgins in Lebanon. August 1989.
12. Invasion of Panama. December 1989.
13. Immediate U.S. response to Iraqi invasion of Kuwait. August 1990.
14. Larger U.S. deployment to Middle East in response to Iraqi invasion of Kuwait. August
1990.
15. First Gulf War begins (air war). January 1991.
16. First Gulf War begins (ground war). February 1991.
17. Military exercises conducted in Kuwait and the Persian Gulf to get Iraqi compliance with
weapons inspections. July 1992.
18. 200 Air Force and Navy aircraft used to enforce "no-fly zone" in Southern Iraq.
September 1992.
19. 30,000 American troops, carrier group deployed in Somalia to facilitate famine relief.
December 1992.
20. Troop deployed in Kuwait and aircraft and missiles used to attack Iraqi military
installations in January 1993.
21. Additional troops, aircraft carrier deployed to Somalia in October and November after
U.S. soldiers killed in October 1993 clash with Somalis.
22. Military exercises in Caribbean simulate an invasion of Haiti. July 1994.
23. 20,000 troops occupy Haiti after agreement with military regime on September 1994.
24. Large ground force, ships, aircraft sent to Persian Gulf region in response to Iraqi threats
to Kuwait. October1994.
25. Carrier task force, Marine contingent, attack submarine, and other ships move into
Adriatic on May 29-30 after UN observers taken hostage by Serbs in Bosnia. May 1995.
26. Troops, ships deployed to Persian Gulf region in response to Iraqi threats in August 1995.
27. Troop deployment to Bosnia as part of Dayton Agreement begins in December 1995.
28. Cuba shoots down American civilian plane. February 1996.
29. The US military launched cruise missile attacks against 14 Iraqi air defense bases
following Iraq's invasion of the Kurdish "safe haven." September 1996.
30. Troops mobilized; b-52s, patriot missiles deployed near Iraq in response to Kurdish area
invasion and inspection violations. September 1996.
31. Iraq ceases cooperation with UN inspectors. October 1997.
32. Iraq expels UN inspectors. November 1997.
33. Clinton threatens major attack on Iraq. February 1998.
34. Operation Infinite Reach (OBL retaliation) - Cruise missile strikes against Afghanistan
and Sudan in response to bombings of two U.S. Embassies in Africa. August 1998.
35. Operation Desert Fox: Attacks on Iraq for inspections violations. November 1998.
36. Iraq orders UN inspectors to leave (again). December 1998
37. Kosovo Air Campaign. March 1999.
38. Bombing of U.S.S. Cole in Yemen- October 2000.
39. Chinese air force forces down US reconnaissance plane- April 2001
40. Initial deployment of troops to Afghanistan- September 2001
41. Afghanistan invasion. October 2001.
42. Second Gulf War. March 2003.
EXPERT SURVEY RESULTS
ONLINE CODING FORM